lts5 / cmp Goto Github PK

does fslmath work correctly?
otherwise save with nibabel

for connectionmatrix, save memory

see also nipy code

what is this directory good for?

check again with the matlab script.
ask ale's neurologist?
difference between the LUT and aseg.stats version?

we provide the different scales. these have to be included as parameters. it should be possible to have different scales, i.e. parcellations to run.
this implies that it has to be included in the configuration. idea:

• provide a graphml file edgeless that contains nodes that represent the intensitity values in the volumes, and additional information (node name, url, color etc.)
• need to add parcelated atlas (freesurfer?). do we need volume based, and/or surface based?

change format of old pipeline format. use data from connectome file converter to know the brain region mappings.

after each module execution, send an email to the specified receipient.

update it in raw_step1.py

it takes too much memory?
is this a fs 5.0 issue?
(tested with diff_unpack)

have fibers read line by line for endpoint creation

check that we missed nothing: e.g.
what about ordred_3parcels.mat / new_order_mai.mat ?

info, error

Apply the REGISTRATION TRANSFORM to the output of FreeSurfer (WM+GM)"

step4_registration_applyToTractoMasks.py and fillInHoles.matlab

What does it do?

Alessandro
In my trials, I found that non-linear registration can help a lot, since many EPI distortions occur very close to the frontal part of the CC. If not properly corrected, these distortions can lead to lots of fibers lost.

Actually, computation time incresases a lot depending in the case...maybe what we can do is first correct the images for EPI distortion and then register with linear registration...do you think that may work?

Alessandro
The problem is "how to correct EPI distortions?" I mean, if you have some field-maps you can do that (actually, I don't know how...but it's possible). In our case I used the registration itself by exploiting an additional T2 image

[email protected]
there's a paper I found in ISMRM... Wu, MICCAI 2008 ... it seems quite good...but still neither I i dont now how to correct the EPI distortion in a kind of 'automatic way'

[email protected]
And talking about EPI and some other artifacts. Has I'm working with baby images, most of the time I have big issues with the movement of the patient...(black hole in the mosaic). For avoiding this, there's a software apparently really good...RESTORE who performs a robust Tensor Fitting to avoid Goshting in images of mosaic. Is from the group of Carlo Pierpaoli. I havent got time to check it out, but maybe we can consider the idea of having a look to it and find if we can profit

[email protected]
DIFF_PREP - software for image resampling, motion, eddy current distortion and susceptibility induced EPI distortion corrections, and for re-orientation of data to a common space...is from Carlo Pierpaoli group...literally, it is a: 'software for image resampling, motion, eddy current distortion and susceptibility induced EPI distortion corrections, and for re-orientation of data to a common space'.
CHECK IT OUT IN: http://www.nitrc.org/projects/tortoise/html.ie6

using sphinx.
who does what?

wrap viewer. where to add invocation?

using dtk

is this an issue?

These datasets MUST BE in the space/orientation of the 'T1' dataset, and after ######
the script they'll be in the space/orientation of 'DSI_b0_resampled' dataset. ######

based on FA value threshold?

to generate automatically after the processing,
e.g. network statistics, variablity in subjects, between timepoints

what we would like to support:

• DSI -> ADC, Kurtosis, GFA (already there)
• q-ball -> ADC, GFA
• DTI -> ADC, FA ( if possible D|| lambda1 and lambda2+lambda3/2, Dorthogonal)

know which module was already processed.
successfully or not.
check output files existing. (rename temporary file trick)

e.g. for face-removed anonymized datasets

byte-code. you can bug-report. if you want to contribute a new module, behaviour. ask for collaboration and grant access to the code (without the core).

http://www.huyng.com/archives/python-logging-from-multiple-processes/418/

based on fiber properties:

• symmetry idea
• torsion, curvature ?

as a thin layer on top of the mapping script

i.e. do not rely on correctly set environment variables in the environment
just use them to find likely correct paths

write it using cfflib in the end.
generate meta.cml.

check with ale if for non-linear registration mode the matrix copying is correct.
use 'T1-TO-T2.mat'?

after discussion with alia, we should consider removing the automated wm mask correction based on gfa parameters. users should not be forced to use this approach.

based on patrics script (stored in scratch/rsfmri) we could produce functional correlations.
but first we need a processing module for rsfmri data:
register fmri to b0, since we have roi to b0, we can compute roi-based averages for time series generation.

in the configuration of the pipeline.
goal: no hard-coded parameters for executable invocation.

the reference is LONI MiND: Metadata in NIfTI for DWI in NeuroImage,Homepage. If we could adopt this standard, I think we will be in good company.

In addition to the SVN checkout, you need to provide the cube1mm.img images

Alessandro, me and [email protected]:

I proposed a small trick for that. I used the FA map to include in the WM mask all the misclassified voxels; for instance, I included all voxels labeled as "GM" and with FA>0.3, because actually they probably are "WM" instead. Then, I checked that the boundary between WM-GM matched as before. This trick helped, but of course THIS IS NOT "the solution".

[email protected]
great...FYI in case of children and newborn this threshold must be reduced at least at FA>0.1 :P the FA values are much smaller than the case of adult brain

Alessandro, me and [email protected]:
Ok. So we could make it parametric allowing people to specify it and interactively see the results! :)

[email protected] and me:
yes, I think is the best think to do...and is not so difficult, I guess...I know that i'm the only one that works with children and newborns, but giving the possibility of 'personalize' this parameter can be usefull for all of us

[email protected]:
The FA value (0:0.1, 0:1, 0:10, 0:100, 0:1000) depends on the version of DTK (diffusion toolkit). I have different data with different FA values depending on what version of DTK I used for the processing.

linear transformation matrix looks quite different!
fs 5.0 issue?
no log output. why?

do we need this checking?

not rely on environment variables setting!
check trackfile headers

Patric: Just thinking about the pipeline, I was thinking that at the same time we are integrating the new freesurfer we should add the insula in the connection matrix, which is really a missing part in the current status. Insular would be one ROI at level 66 and be many randomly organized ROIs at level 1000. For the intermediate, we just need to group iteratively in a way that ROIs are as compact as possible. If you need any help for the grouping just let me know. For the generation of the small ROIs use Xavier's partition process.

like format of diffusion toolkit (dsi_odf + odf_max) ?

check again with the matlab script.
ask ale's neurologist?
difference between the LUT and aseg.stats version?

from mri_info, stable5

even the same mri_convert syntax is applied as elsewhere

[email protected],
Well here I suppose that it would be better on the registered T1 if you have a whole brain coverage of the diffusion data, which is my case (44 slices). But of course if you have data where some slices are missing (in Alessandros case), it's better to do it the other way. This could be something we add to the new pipline as an option (whole brain do this if not do that.....)

I would like to have the possibility to define any kind of cortical and subcortical parcellation in the pipline. There are some efforts to improve parcellation:
www.braincolor.org

ROI files don't match between different resolutions: if you overlap them, you can see they differ in several regions

1. extract surface mesh as graph
2. extract one region as subgraph
3. apply graph partitioning: http://mathema.tician.de/software/pymetis
4. do subregions be adjacent enumerated?
5. put it together again

• It could be useful to have only one folder for each subject, which contains all the that the pipeline worked on (e.g. Freesurfer data, cmt data etc). I found annoying to have the data sparse in different parts of the hard-disk.

I think it can be solved, for example, by having a variable in which one can specify the project he is working on (stroke, epilepsy etc). Then, we will have a big folder containing one subfolder-per-project. Inside this folders, there will be subfolder for patients data, and inside these we could have three folder:

• rawdata: which will contain all the DICOM rawdata, the one converted to NIFTI and all the registration stuff (additional data (e.g. T2), tranformations matrix, warping etc ).
• freesurfer: which will contain all data produced by freesurfer. I think this would only require to set the variable $SUBJECTS appropriately.
• cmt: which will contain all data produced by the pipeline (fsmask_1mm, ROI_HR_th, fiber-tracking data files, connection matrixes etc). I think this would only require to set the variable $CMT_SUBJECTS appropriately

have 1 or 2. is the first always 0?